As the internet is widely used, access to useful information is becoming easier but side effects concerning the distribution of an overflow of harmful information are also increasing. Easy access to harmful information by young children who need social protection and people who lack self-control has already become a social problem.
Harmful information may be classified into text, images, sound, multimedia contents, etc. Considering the share of harmful images in the harmful information and the language-independent features of image information, great importance is placed on the research of the technology for harmful image classification, identification and nudity analysis, which is an image-based harmfulness determination technology, among harmful information analysis technologies.
N. Fleck et al.'s “Finding Naked People”, In European Conf. on Computer Vision, Vol. II, 1996 proposed a content-based image retrieval (CBIR) using the technology of image feature extraction and classification. Since then, there have been many researches on extraction of features contributing to a harmful image and analysis of nudity of the image using the extracted features and a learning-based identification algorithm.
Elemental technologies of the harmful image classification technology include a harmful image preprocessing technology, a harmful image feature extraction technology, a harmful image identification technology, a harmful image classification system integration technology, a harmful image classification system evaluation technology, and so on. Among these, the harmful image feature extraction technology and the harmful image identification technology have been mainly studied. These technologies are used in many areas, including internet-based multimedia services using multimedia contents, such as web portals, user created contents (UCC), peer-to-peer (P2P), web hard, etc, broadcast contents, movie contents, and so forth.
Most of the conventional technologies for harmful image classification, identification and nudity analysis employ a method using a hash or meta database. However, this method has a limitation because the method involves analysis using determination values of harmfulness or harmlessness of an entire image based on only parts of the image. Moreover, it is difficult to apply the SafeNet standards of Korea Communications Standards Commission, which are the national standards on harmful contents.